According to Lin & Luo (2015, p. 2404) robot welding system usually consists of “robot manipulator, welding power source, welding torch, wire feeder, positioner, welding torch cleaning and calibration station, fume extraction, and safety fence” and seam tracking device. Robot station configuration can have both stationary or moving robot and a workpiece. The movement can be realized with a column, a gantry or a track. Multi-robot welding system is typically used when high productivity is wanted or the size of the workpiece requires multiple robots to weld. Multiple robots can be used for welding simultaneously or one of the robots can be used for handling of the workpiece. (Lin & Luo 2015, p. 2404–2406.)
According to Vuong, Lim and Yang (2015) traditional way to make robot welding programs has been the walk-through programming and lead-through programming. These programming techniques are called on-line programming and they require industrial robot to physically to move to the target locations, either by operator guiding the robot with a joystick attached near the gripper (walk-trough) or with a teach pendant (lead-trough). This means that every time a robot program is made, a downtime to the production process is evitable. (Vuong, Lim and Yang 2015, p. 2072–2073.) To avoid production downtime many robotic welding simulation software have been developed. Robotic welding simulation software have tools to make robot programs from simulations, which can be called as an off-line programming. The 3D model of the welding robot cell can be created with robot simulator software. The work cell model and computer aided design (CAD) model of the workpiece can be used to generate the geometric information needed in robot program, such as target points for robot paths. The robot program is generated by combining the geometric information and robot kinematic/dynamic model information. It is worth mentioning that even though the robot simulations try to replicate the actual working environment as accurately as possible, there will always be some differences between the real world and a simulation. These differences can cause several unwanted problems, such as a collision of a robot. (Lin & Luo 2015, p. 2437–2438; Vuong, Lim and Yang 2015, p. 2073–2075.) Generally the manufacturing task with industrial robot can be expressed as a robot programming process in following five steps (Vuong, Lim and Yang 2015, p. 2075–2076):
dividing of the objective into sub objectives
breaking of each sub objectives into simple instructions or commands, which can be executed by the robot controller
gather geometric information for movement instructions
set process parameter information for the manufacturing task
combine geometric information and process parameters to form a robot program.
The first and second step is carried out by human as they require noticeable amount of intellectual capability, which will be problematic for current state of robot intelligence.
Therefore, the scientific research to improve robot programming has mainly focused to steps 3 and 4. (Vuong, Lim and Yang 2015, p. 2075–2076.)
Robots in the welding cell must avoid collision with each other and with the workpiece.
Therefore, motion planning is needed in creating of trajectories free of collision. The motion planning of multi-robot welding has been widely researched. Pellegrinelli et al. (2017) proposed an approach where cell design and motion planning problems are solved simultaneously. Chao & Sun (2017) proposed a theory of multi-robot motion planning which uses genetic algorithm.
According to Pellegrinelli et al. (2017) the common techniques in solving the motion planning problem are: “potential fields, roadmaps, cell decomposition, probabilistic potential fields, probabilistic roadmaps, probabilistic cell decomposition and simple-query sampling-based method.” Two common methods have been developed to solve the motion planning problems the first one is called decoupled planning and the second centralized motion planning. In decoupled planning the movement of every robot is defined one at the time and the existence of other robots is ignored. After the movement for each robot is defined, the paths are combined and collisions between the paths are resolved by adjusting the velocity of the robots or by changing the robot path. In centralized motion planning all of the robots in the welding cell are considered as a one operating multi-body robot. This creates higher dimensionality of the configuration space than in the decoupled planning, but according to Sanches & Latombe (2002) centralized planning has shown to be more efficient than decoupled planning. (Pellegrinelli et al. 2017, p. 99.)
In a research by Chao & Sun a genetic algorithm is used as a basis for creating an approach for collision free multirobot motion planning of spot welding robots. Genetic algorithm can be used to optimize the welding sequences and to minimize the total welding time. The algorithm functions as shown in figure 10, first the constraints are given, such as number of robots, number of welds, welding time and sequence constraint, then the initial order of welds are given for each robot and finally the genetic algorithm calculates the optimal welding sequence for the welding robots. (Chao & Sun 2017, p. 193–201.)
Figure 10. Illustration of how genetic algorithm can be implemented to multirobot welding (Chao & Sun 2017, p. 195).
As the genetic algorithm only produces the optimal welding order and welding path, simulation is required to ensure that no collisions occur during welding. If collision occurs, the robot position can be modified and if collision free position is still not found, the welding sequences can be changed, which also changes the robot path. Otherwise a new optimization iteration of robot paths with genetic algorithm is required to find the collision free paths.
(Chao & Sun 2017, p. 193–201.)